So what do you think? How real is that risk? And also, how do we avoid that happening?

It is real.

It's already being measured.

And it's a design problem.

Not a reason to start, though.

So remember, I'm a gastroenterologist.

There's a multi-center colonoscopy study where we cite where endoscopists who'd been working with AI-assisted polyp detection got measurably worse at finding polyps when they went back to working without it.

How do we make sure clinicians retain that expertise to make them good human clinicians, not just nice humans who practice medicine?